RESEARCH ARTICLE Open Access

Comparison of immunohistochemistry withPCR for assessment of ER, PR, and Ki-67and prediction of pathological completeresponse in breast cancerHans-Peter Sinn1*, Andreas Schneeweiss2, Marius Keller1, Kornelia Schlombs6, Mark Laible6, Julia Seitz2,Sotirios Lakis4, Elke Veltrup4, Peter Altevogt3, Sebastian Eidt5, Ralph M. Wirtz4,5 and Frederik Marmé2

Abstract

Background: Proliferation may predict response to neoadjuvant therapy of breast cancer and is commonlyassessed by manual scoring of slides stained by immunohistochemistry (IHC) for Ki-67 similar to ER and PgR.This method carries significant intra- and inter-observer variability. Automatic scoring of Ki-67 with digital imageanalysis (qIHC) or assessment of MKI67 gene expression with RT-qPCR may improve diagnostic accuracy.

Methods: Ki-67 IHC visual assessment was compared to the IHC nuclear tool (AperioTM) on core biopsies from arandomized neoadjuvant clinical trial. Expression of ESR1, PGR and MKI67 by RT-qPCR was performed on RNAextracted from the same formalin-fixed paraffin-embedded tissue. Concordance between the three methods(vIHC, qIHC and RT-qPCR) was assessed for all 3 markers. The potential of Ki-67 IHC and RT-qPCR to predictpathological complete response (pCR) was evaluated using ROC analysis and non-parametric Mann-Whitney Test.

Results: Correlation between methods (qIHC versus RT-qPCR) was high for ER and PgR (spearman´s r = 0.82,p < 0.0001 and r = 0.86, p < 0.0001, respectively) resulting in high levels of concordance using predefined cut-offs.When comparing qIHC of ER and PgR with RT-qPCR of ESR1 and PGR the overall agreement was 96.6 and 91.4%,respectively, while overall agreement of visual IHC with RT-qPCR was slightly lower for ER/ESR1 and PR/PGR(91.2 and 92.9%, respectively). In contrast, only a moderate correlation was observed between qIHC and RT-qPCRcontinuous data for Ki-67/MKI67 (Spearman’s r = 0.50, p = 0.0001). Up to now no predictive cut-off for Ki-67assessment by IHC has been established to predict response to neoadjuvant chemotherapy. Setting the desiredsensitivity at 100%, specificity for the prediction of pCR (ypT0ypN0) was significantly higher for mRNA than forprotein (68.9% vs. 22.2%). Moreover, the proliferation levels in patients achieving a pCR versus not differedsignificantly using MKI67 RNA expression (Mann-Whitney p = 0.002), but not with qIHC of Ki-67 (Mann-Whitneyp = 0.097) or vIHC of Ki-67 (p = 0.131).

Conclusion: Digital image analysis can successfully be implemented for assessing ER, PR and Ki-67. IHC for ERand PR reveals high concordance with RT-qPCR. However, RT-qPCR displays a broader dynamic range and highersensitivity than IHC. Moreover, correlation between Ki-67 qIHC and RT-qPCR is only moderate and RT-qPCR withMammaTyper® outperforms qIHC in predicting pCR. Both methods yield improvements to error-prone manualscoring of Ki-67. However, RT-qPCR was significantly more specific.

Keywords: Image analysis, Breast cancer, Ki67, mRNA, RT-qPCR, Prediction, Pathologic complete response,neoadjuvant, Immunohistochemistry (IHC), MammaTyper®

* Correspondence: peter.sinn@med.uni-heidelberg.de1Institute of Pathology, University of Heidelberg, Im Neuenheimer Feld220-221, 69120 Heidelberg, GermanyFull list of author information is available at the end of the article

© The Author(s). 2017 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, andreproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link tothe Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Sinn et al. BMC Cancer (2017) 17:124 DOI 10.1186/s12885-017-3111-1

BackgroundThe proliferative activity of individual cells is a hallmarkof tumor biological aggressiveness and a key determinantof sensitivity to (neo)adjuvant chemotherapy, thus beingamong the principal factors guiding clinical managementin primary breast cancer [1, 2]. The most widely usedmethod to assess proliferation as well as hormone recep-tor expression is immunohistochemistry (IHC).Nuclear staining of the nuclear antigen Ki-67 is most

widely used as a surrogate for proliferative activity. Ki-67 is present in the cell nucleus throughout all stagesof the cell-cycle excluding the resting phase G0 [3].The recently proposed St Gallen recommendations forthe identification of the intrinsic subtypes using surro-gate pathologic-based definitions have underlined thevalue of Ki-67 as a clinical tool in routine clinical prac-tice [2]. Ki-67 is recommended as a valuable factor todistinguish between Luminal A- and B-like tumors, afundamental distinction in clinical decision-makingtoday. [4–6].Despite its widespread use, driven by the premise of

solving delicate therapeutic dilemmas combined withseveral advantages such as universal accessibility, easyapplication and low cost, the assessment of Ki-67, ERand PR is affected by technical and observer-basedvariabilities of the IHC method [7, 8]. This can be illus-trated by observations, that tumors with as little as 1%positive nuclei still respond to anti-hormonal treat-ment, which indicates that tumor cells lacking nuclearER staining within in these tumors do have someextend of ER expression rendering them sensitive to es-trogen deprivation or estrogen receptor blockade [9].While the clinical role particularly of ER testing by IHCis well established, the clinical utility of Ki-67 is stillcontroversial [10]. The reason lies in a series of analyticaland preanalytical factors, but also in staining interpret-ation and scoring [11]. Importantly, attempts to reducethe high discordance rates either by means of formalcounting quantification methods (as opposed to simpleeyeballing) or by training of individuals have not beensuccessful [12].Despite methodological concerns, overall a strong

correlation of Ki-67 with breast cancer outcome is suf-ficiently supported, particularly by data originating fromrandomized clinical trials with central review of bio-markers [10]. This has also been shown in the neoadju-vant setting, where higher Ki-67 values are consistentlyassociated with higher rates of pathological complete re-sponse (pCR) [13], a finding which reflects the fundamen-tal link between tumor replication fraction and activity ofcytotoxic agents. Still it remains difficult to identify a rea-sonable cut-off to predict pCR [14].Two techniques which could circumvent the inter- or

intra-observer variability of Ki-67 manual microscopic

assessment are automated image analysis and reversetranscription quantitative real-time PCR (RT-qPCR). Atrained human eye may achieve an excellent under-standing of images and patterns, but is less accuratewhen it comes to quantification. Computer-based vi-sion methods could represent a solution to this prob-lem by offering standardized image processing andreliable quantification [15]. However with regard to theassessment of tumor proliferation, the areas of interestand staining intensities have to be defined, and mea-sured reproducibly. Also, it is still under debate how todeal with areas of increased proliferative activity (hotspots), and if low intensity staining should be taken intoaccount [16]. Therefore, automated estimations of Ki-67 are highly correlated with manual assessments, butit is not yet certain whether or not they can improveprediction and prognostication [17–19].RT-qPCR has a series of widely acknowledged meth-

odological advantages over IHC, which appear particu-larly beneficial in the context of reducing the bias ofroutine Ki67 assessment; it is quantitative by nature withmuch wider dynamic range, it does not require an expe-rienced eye, and results are not affected by subjectiveinterpretations [20]. Moreover, access to standardizedprotocols and automation ensures accurate performanceand fast turn-around. In recent years highly specific andsensitive techniques have been developed, which allowfor fast and efficient extraction of high-quality nucleicacids from FFPE overcoming the challenges posed byfixation and embedding [21].Validation of the various available methods for the

assessment of Ki-67 requires comparative testing prefera-bly in a specifically defined clinical context. In the presentstudy we used the neoadjuvant setting of a phase II trialrandomizing patients receiving anthracyclin/taxane basedstandard treatment between pemetrexed and cyclophospa-mide in order to directly compare the assessment of Ki-67with automatic, quantitative read-out of IHC (qIHC) andthe determination of tumor MKI67 mRNA with RT-qPCRon FFPE tissue extracted RNA.

MethodsStudy populationCore needle biopsies from 101 out of 105 patients (96,2%)with primary invasive breast cancer, that had beenenrolled in the H3E-MC-S080 (NCT00149214, Sponsor:Eli Lilly and Company) neoadjuvant phase II study [1],were obtained. All patients had been diagnosed with oper-able (T2-T4/N0-2/M0) breast cancer at a single institution(National Center for Tumor Diseases, University-Hospital,Heidelberg) had been randomized to receive sequentialanthracycline/taxane-based regimens containing eitherpemetrexed or cyclophosphamide in combination withepirubicin. A written informed consent for the research

Sinn et al. BMC Cancer (2017) 17:124 Page 2 of 10

use of patient biological material was granted at the timeof enrolment. The study was approved by the local ethicscommittee. Complete molecular data (including RT-qPCRdata) and clinical follow-up information were available in83 out of 105 (79%) patients (statistics data set #1). Ki-67,ER, and PR IHC slides were available in 54 (51%) patientsfor quantitative IHC (statistics data set #2).

Isolation of tumor RNAFor RNA extraction from FFPE tissue, a single 10 μm curlwas processed according to a commercially availablebead-based extraction method (RNXtract® kit; BioNTechDiagnostics GmbH, Mainz, Germany). In brief, a lysisbuffer was used to liquefy FFPE tissue slices while meltingof paraffin was carried out in a thermo-mixer. Tissuelysis was accomplished with a proteinase K solution.Thereafter, lysates were admixed with germanium-coated magnetic particles in the presence of specialbuffers, which promote the binding of nucleic acids.Purification was carried out by means of consecutivecycles of mixing, magnetization, centrifugation andremoval of contaminants. RNA was eluated with 100 μlelution buffer and RNA eluates were then stored at−80 °C until use.

Gene expression by RT-qPCRThe MammaTyper® is a molecular in vitro diagnostictool for the assessment of the gene expression levels ofthe four cancer biomarkers that are required for theclinical management of breast cancer patients in dailyroutine clinical practice. Instead of using IHC to assessprotein expression of HER2, ERα, PR, and Ki-67, withMammaTyper®, it is possible to measure the mRNAtranscripts of the corresponding genes (ERBB2, ESR1,PGR, and MKI67), doing so by using routine FFPEmaterial and by achieving accurate, reproducible andobjective results. The gene expression data may be thenintegrated so as to assign individual samples to a mo-lecular subtype of breast cancer.The mRNA expression levels of ERBB2, ESR1, PGR,

and MKI67 as well as of two reference genes (REF),namely B2M and CALM2, were determined by RT-qPCR, which involves reverse transcription of RNA andsubsequent amplification of cDNA executed successivelyas a 1-step reaction. In MammaTyper®, the 6 assays(assay = primer pair and probe specific for the respectivetarget sequence) are duplexed into three assay mixes,each using a pair of hydrolysis probes labelled with differ-ent fluorophores for separate detection of the duplexedassays [22].Each patient sample or control was analyzed with each

assay mix in triplicates. The experiments were run on aVersant kPCR Molecular System (Siemens Healthcare,Erlangen, Germany) according to the following protocol:

5 min at 50 ° C, 20 sec at 95 ° C followed by 40 cycles of15 sec at 95 ° C and 60 s at 60 ° C and according toMammaTyper® instructions for use 140603-90020-EURev 2.0.Forty amplification cycles were applied and the cycle

quantification threshold (Cq) values of MKI67 and thetwo REF genes for each sample (S) were estimated as themedian of the triplicate measurements. These were thennormalized against the mean expression of the REFgenes and set off against a calibrator (PC), to correct forinter-run variations (ΔΔCq method) (Livak et al. 2001).The final values were generated by subtracting ΔΔCqfrom the total number of cycles to ensure that normal-ized gene expression obtained by the test is proportionalto the corresponding mRNA expression levels, a methodthat facilitates interpretation of data and clinicopathologi-cal correlations. The various calculation steps are summa-rized in the following formula:

40‐ΔΔCq MKI67ð ÞS ¼ 40‐ð Cq MKI67½ �S – meanCq REF½ �Sð Þ– Cq MKI67½ �pc – meanCq REF½ �pcð ÞÞ

In 18 patients the MammaTyper® assay failed, becausethe required amount of RNA was not sufficient foranalysis according to pre-specified criteria as describedin the instructions for use.

Pathology and ImmunohistochemistryTumor grading, tumor typing and immunohistochem-istry (ER, PR, Ki-67) was performed on the pretreat-ment core biopsies on all patients. Pathologicalcomplete response (pCR) was determined on tumorresection specimens after completion of neoadjuvantchemotherapy, and was defined as no evidence of re-sidual invasive and ductal disease in the breast andlymph nodes (ypT0,ypN0).Immunohistochemistry was performed according to

previously standardized protocols on an automated IHCplatform (Dako Techmate 500) with citrate buffer forantigen retrieval [23] and observing the ASCO/CAPguidelines for immunohistochemistry [7]. The followingprimary antibodies and corresponding dilutions wereused (DakoCytomation, Glostrup, Denmark): ER (clone1D5, 1:100), PR (clone PgR636, 1:100) and Ki-67 (MIB-1,1:200). Slides were assessed by quantitative image ana-lysis (qIHC) using the Aperio Image Analysis toolbox(Leica Biosystems, Nussloch, Germany). Staining inten-sity and percentage of positive nuclei were recorded aftermanually segmenting tumor from adjacent stroma.Tumors with ER/PR Remmele scores greater than 3 orpositive nuclei greater than 1% were considered hor-mone receptor positive.

Sinn et al. BMC Cancer (2017) 17:124 Page 3 of 10

Statistical methodsThe Spearman correlation coefficient r was used as ameasure of the strength and direction of the linear rela-tionship between variables. 2×2 contingency tables wereused to calculate positive percent agreement (PPA) andnegative percent agreement (NPA) as a measure of agree-ment: PPA = 100% x a/(a + c), NPA = 100% x d/(b + d).Receiver Operating Characteristics (ROC) analysis wasperformed to determine the optimal cut-off for Mamma-Typer® gene and qIHC protein measurements with pCRas the endpoint. ROC analysis instead of comparing oddsratios to take into account the ratios of clinically relevantfalse positive and false negative determinations and toidentify cut points for each method at clinically relevantprerequisites (i.e., detect all responding tumours). ROCanalysis has been used to objectively address each methodproviding different result codings in a non-parametricmanner [24]. On the other hand ROC analysis bears therisk of misinterpreting clinical validity when analyzingheterogeneous populations [25]. However, we have ana-lyzed the response to neoadjuvant chemotherapy withincontrolled, randomized phase II trial which has definedinclusion and exclusion criteria to have the most compar-able basic risk situation. As optimal cut-off for the identifi-cation of complete response by the methodologies thepoint of highest sensitivity still retaining 100% specificitywas chosen. The p value reported for evaluating the ROCcurve tests the null hypothesis that the area under thecurve really equals 0.50 as provided by the statistical pro-gram used (GraphPad Prism). The non-parametric Mann-Whitney test was used to confirm the statistical signifi-cance when comparing responding versus non-respondingtumors and box plots were used to illustrate each case ofresponding and non-responding tumor above and belowthe cut-off value. Statistical analyses were performed withJMP SAS (SAS Institute, Cary, NC, USA) and Graph PadPrism software (Version 5.04; Graph Pad Software Inc., LaJolla, CA, USA).

ResultsPatient populationBiopsy tissue was available from 101 out of 105 patients.Gene expression analysis by MammaTyper® was successfulin 83 biopsy specimens with full clinical data out of a totalof 105 trial participants (Fig. 1). 12 patients out of thisgroup had achieved complete pathological remission(pCR). Basic clinicopathological characteristics of statisticsdata set #2 that includes quantitative IHC data is listed inTable 1.

Comparison of IHC with RT-qPCR based assessment of ER,PR and Ki-67Comparing qIHC of ER with RT-qPCR of ESR1demonstrated a good overall agreement of 96,6% (PPA

100%; NPA 92.3%) as well as a good correlation looking atthe continuous data (spearman’s r = 0.82, p < 0.0001).Correlation between vIHC of ER and RT-qPCR for ESR1(spearman’s r = 0.85, p < 0.0001) and between vIHC andqIHC ER (spearman’s r = 0.88, p < 0.0001) was high, too.

Fig. 1 Remark diagram of sample selection

Table 1 Basic tumor characteristics

no-pCR pCR

Histological Type p= 0.39

Invasive ductal NOS 37 82.2% 9 100.0%

Invasive lobular 7 15.6% 0 0.0%

Other 1 2.2% 0 0.0%

Histological Grade p= 1.00

Grade 2 15 33.3% 3 33.3%

Grade 3 30 66.7% 6 66.7%

ypT Category p= 0.81

ypT0 0 0.0% 9 100.0%

ypT1 27 50.0% 0 0.0%

ypT2 9 20.0% 0 0.0%

ypT3 8 17.8% 0 0.0%

ypT4 1 2.0% 0 0.0%

ypN Category p= 0.81

ypN0 27 50.0% 9 100.0%

ypN+ 25 46.3% 0 0.0%

ypNX 2 3.7%

HER2 status p= 0.68

Neg 36 66.7% 7 13.0%

Pos 7 13.0% 2 3.7%

Sinn et al. BMC Cancer (2017) 17:124 Page 4 of 10

Overall agreement for PR protein and PGR mRNA ex-pression was 91.4% (PPA 83.3%; NPA 100%) comparingqIHC and RT-qPCR and there was a high correlation forthe continuous data (r = 0.86, p < 0.0001). Correlationbetween vIHC and RT-qPCR and between vIHC andqIHC was very high, too (r = 0.88, p < 0.0001 and r = 0.90,p < 0.0001, respectively). Concordance when comparingvisual IHC protein with RT-qPCR RNA expression wasgood for ESR1 as well as for PGR, although slightly lowercompared to the agreement between qIHC and RT-qPCR(OPA 91.2%; PPA 90.9%; NPA 91.7%) and for PGR (OPA92.9%; PPA 88.0%; NPA 96.9%, respectively). For both,ESR1 and PGR, only 4 cases were discordant, with 3 caseseach positive by vIHC and negative by RT-qPCR, while 1case was negative by vIHC and positive by RT-qPCR(Fig. 2). However, several of these discrepancies could beresolved by using quantitative IHC, as these cases werealso discrepant when comparing vIHC with qIHC. More-over, qIHC could delineate quantitative differences of

hormone receptor expression at the highest RemmeleScore value of 12, where vIHC could not resolve expres-sion differences. In addition, at the lower range of ex-pression levels RT-qPCR based assessment could stilldetermine substantial differences of mRNA levels whilethe IHC based assessment could not detect any proteinexpression. The inter-gene spearman correlation wasmoderate for ESR1 and PGR (r = 0.59, p < 0.0001), whileKi-67 correlated negatively with PGR (−0.37, p = 0.007).While the correlation for ESR1 and PGR protein andRNA expression was high when comparing IHC with RT-qPCR results (spearman’s r = 0.82, p < 0.0001 and r = 0.86,p < 0.0001, respectively), the correlation between MKI67protein and RNA expression was only moderate (spear-man’s r = 0.56 for vIHC and r = 0.47 for qIHC). In contrastthe correlation between both methods for Ki-67 assess-ment by IHC was high (r = 0.80, p < 0.0001). The medianvalue of Ki-67 proliferation index by image analysis IHC(qIHC) was 23.4%, by conventional visual IHC (vIHC)

Fig. 2 Correlation of RT-qPCR for ESR1, PGR and MKI67 with quantitative IHC by image analysis (a, c, e) and visual IHC assessment (b, d, f)

Sinn et al. BMC Cancer (2017) 17:124 Page 5 of 10

35.0%, and 37.01 for RT-qPCR, clearly reflecting the inclu-sion criteria of the S080 trial which targeted clinicallyhigher-risk patients. Scatter plot analysis displays the posi-tive correlation between RT-qPCR and visual as well asquantitative IHC assessment in Fig. 4.

Prediction of pCRTo compare the clinical utility we performed a ROCanalysis to determine the optimal cutoff for predictingthe pCR. The results of the ROC analysis are presentedin the graphical plots of Fig. 3. With RT-qPCR, 100% ofresponders could be detected with a specificity of 68.9%at a 40-ddCT level of 37.31 which almost reflected themedian mRNA expression in this cohort (Fig. 3a, b).Conversely, no responder was below RT-qPCR of 37.31(Fig. 4a). For IHC assessment, it was difficult to deter-mine a reliable cutoff reaching high sensitivity and speci-ficity. With RT-qPCR the area under the curve was 0.78for the overall cohort and 0.80 for the IHC cohort(Statistics #2) (p = 0.002 and p = 0.004). For both IHCmethods, the ROC was not significant.

Since maximum sensitivity was a pre-requisite, thetwo methods were compared with respect to specifi-city, which was found to be substantially higher forMammaTyper® (68.9%) compared to qIHC (22.2%).Using the cut-offs indicated by ROC analysis (37.31for RT-qPCR, 13.2% for qIHC and 3.5% for vIHC),tumors were characterized as bearing either high orlow MKI67 RNA or Ki-67 protein expression, respect-ively. However, as illustrated in Fig. 4, statisticallysignificant differences between groups were found forthe RT-qPCR but not for IHC methods when patientswere stratified according to proliferation and pCR(Mann–Whitney p = 0.003 and p = 0,005 for RT-qPCRand p = 0.099 for qIHC and p = 0.133 for vIHC).Using the cut-offs obtained by ROC analysis, pCRwas observed in 9 of 24 patients (37.5%, p < 0.001)with high MKI67 RNA expression but in no patientwith low RNA expression. Accordingly for qIHC, pCRwas observed in 9 of 44 patients (20.5%, p = 0.27)with high Ki-67 labelling index and similarly it wasentirely lacking in patients with low proliferation. ForRT-qPCR the ROC analysis was also highly significant

Fig. 3 ROC analysis for prediction of pathological complete response by quantifying MKI67/Ki-67 expression by RT-qPCR (a, b) and IHC (c, d) showingoverall increased ability of mRNA assessment to correctly identify responders versus non-responders

Sinn et al. BMC Cancer (2017) 17:124 Page 6 of 10

when only luminal tumors had been assessed, though thesample size was small in this subset (data not shown).

DiscussionIn this study we have validated clinical performance ofhormone receptor gene expression by RT-qPCR by com-paring predefined cut-offs in a blinded fashion with thecurrent standard of IHC. Furthermore, we have investi-gated the diagnostic performance of two methods forassessing MKI67 gene expression, namely IHC with com-puterized quantification of protein and RT-qPCR RNAquantification with the MammaTyper® IVD kit in thesetting of pCR prediction. When continuous data were di-chotomized to reflect high- and low-MKI67 categorieswith cut-offs obtained by ROC curve analysis after consid-ering 100% sensitivity, RT-qPCR was significantly morespecific than qIHC.To the best of our knowledge this is the first direct

comparison of this kind in the context of a clinical trial.

For the mRNA estimation we used the MammaTyper®, anovel in vitro diagnostic test for breast cancer molecularsubtyping. To prove the clinical utility of mRNA basedassessment, we compared RT-qPCR with conventionalvisual assessment as well as digital image analysis baseddetermination at a reference pathology lab in the contextof a clinical trial. Moreover, the methods were examinedwith respect to their ability to predict pCR according toKi-67 protein or MKI67 mRNA expression levels mea-sured on pretreatment core biopsies. Our results indi-cate, that, when using RT-qPCR valid cut-offs formRNA expression, which reliably distinguish betweennon-responding and responding tumors as determinedby pCR (ypT0 ypN0) can be identified.Pathological complete response has gained wide ac-

ceptance as one of the strongest predictors of prolongedsurvival in the setting of neoadjuvant chemotherapy[26, 27]. Therefore, laboratory assays that can efficientlypredict a patient’s response to a given preoperativechemotherapeutic combination may serve as tools forindividualizing treatment and improving long-term out-comes [28]. As with adjuvant chemotherapy, neoadju-vant regimens also suffer from the fact that substantialtherapeutic benefit is restricted only to a fraction ofthose treated, whereas all patients will experience ad-verse events because of toxicity [29].In several neoadjuvant studies Ki-67 protein expres-

sion has been investigated in pre-operative biopsies inrelation to the response to treatment and in most casesa high Ki-67 proliferation rate was predictive of higherprobability of pCR [13]. Fasching et al. analyzed Ki-67by conventional IHC in core biopsies from 552 patientsfrom a single German institution and showed that apre-defined 13% cut-off could predict pCR with 94%sensitivity and 36% specificity [30]. Interestingly, ourROC analysis for qIHC requiring 100% sensitivity withthe least possible loss on specificity led to an identicalcut-off (13.2%) for Ki-67. However, this finding requirescareful interpretation, due to differences characterizingthe clinical settings between the two neoadjuvant stud-ies and the original work by Cheang [31]. In the lattercase, the Ki-67 cut-off was fine-tuned against gene ex-pression profiling in order to distinguish Luminal Afrom Luminal B tumors in a population containingboth high- and low-risk breast cancers, whereas in theneoadjuvant setting the same cut-off was intended toidentify the majority of, mainly high-risk, patients thatwould most likely benefit from preoperative cytotoxictherapy.Alike what has been repeatedly shown in the adjuvant

setting, it appears that the molecular architecture of tu-mors as defined by the expression of hormone receptorsand HER2/neu may act as a modifier of the associationbetween Ki-67 and response to neoadjuvant treatment

Fig. 4 Scatter plots illustrating the distribution of RT-qPCR mRNA(upper panel) and qIHC (lower panel) and vIHC (right panel) proteinmeasurements in relation to the groups of responders (green dots)and non-responders (blue dots). Differences were tested with theMann-Whitney test (a = data set 1, n = 83, b, c, d = common dataset, n = 54)

Sinn et al. BMC Cancer (2017) 17:124 Page 7 of 10

and between pCR and long-term outcomes [14, 32].While 101 tumors were available for analysis, the inclu-sion of 83 or 54 tumors in this study was not based on astatistical rational but was dictated by the availability oftumor tissue with complete RT-qPCR and qIHC data.A novel aspect of the present work is the comparison

between protein-based and mRNA-based methods for theassessment of tumor proliferation. Our findings highlightthe feasibility of using RT-qPCR for the routine assess-ment of ESR1, PGR & MKI67 in order to assist the selec-tion of breast cancer patients for neoadjuvant treatment.Even though both RT-qPCR and qIHC of MKI67/Ki-67could be calibrated to maximize negative predictive value,only with the former this was achieved whilst ensuringsufficient specificity, which if validated would signifythat MammaTyper® could help a considerable numberof patients safely forego unnecessary treatment. Thesedata collectively indicate that MammaTyper®MKI67RNA was overall more representative of the true prolif-eration state of the tumor than was computer assistedKi-67 protein estimation, a finding that is worth valid-ating in larger datasets.Significant correlations between conventional Ki-67

visual assessment and RT-qPCR have been previouslyreported [33, 34], indicating a strong biological link be-tween mRNA and protein expression despite methodo-logical variations, as is further indicated by comparableprognostic hazard ratios obtained by both methods[35]. To the best of our knowledge however, our studyis the first to compare image analysis with RT-qPCR forthe assessment of tumor proliferation with the add-itional advantage of using material from a randomizedclinical trial. The correlation between mRNA and pro-tein was significant but moderate, a finding which mayreflect post translational modifications or may be re-lated to the increased dynamic range of RT-qPCR ascompared to IHC. Another possible explanation mightbe that mRNA levels are a reflection of the averagegene expression in the entire FFPE slice, whereas IHCmay be biased in favor of selected “representative”tumor areas. Even in the case of image analysis systems,inspection of digitalized images and manual identifica-tion of tumor areas is necessary before automaticscoring.Computerized methods have been recommended as a

solution to the problem of subjectivity in the visualassessment and scoring of IHC-stained slides. Not sur-prisingly, Ki-67 scores from image analysis systems aregenerally in close agreement with those of manualmethods because manual scoring for research purposesis customarily performed by a pathologist with long-standing experience in the field [17, 36]. It is worthmentioning, however, that in a routine decentralizedsetting, digital processing and scoring of slides would

probably outperform manual assessment which is proneto considerable subjectivity often not improved uponstandardization [37]. Digital analysis yields more repro-ducible results with regard to staining intensity, byfacilitation the definition of low grade staining inten-sities. Definitive conclusions would require compari-sons between all three methods (central versus localversus automatic) performed preferably in the prospect-ive retrospective setting of a large multi-center trial.Multi-gene molecular signatures have also been

tested as a way for predicting pCR in patients withbreast cancer [38–40]. However, generalized use ofthese commercialized assays is limited by their in-creased cost and the requirement to run in centralizedplatforms or both. Interestingly, proliferation genes, in-cluding MKI67, are often heavily weighted in multi-gene scores which serve as estimators of a patients’ riskof developing recurrences. This is perhaps one of thereasons why multi-gene tests do not always prove to beconvincingly superior to conventional or less sophisticatedmethods for tumor risk stratification [35, 41, 42], leadingsome authors to question their cost-effectiveness [43].Moreover, for several commercially available tests, neitherdoctors nor consumers can gain access to the continuousexpression data of individual proliferation markers thatmake up the final risk scores. This restriction overallminimizes the possibility of potentially interesting com-parisons between proliferation motifs or scores and sin-gle proliferation markers based on RT-qPCR or IHC.Strikingly, our ROC curve analysis of MKI67 40-ΔΔCqvalues for the prediction of pCR displayed performancecharacteristics that are comparable with those of a 50-gene predictor of tumor recurrence risk developed bysupervised training of Cox models [39]. Along theselines, single-gene MKI67 RT-qPCR may be worth con-sidering as a golden means for assessing tumor prolifer-ation due to its unique ability to combine technicaladvancements and diagnostic accuracy with more af-fordable pricing.

ConclusionsImage analysis-assisted scoring of ER, PR and Ki-67IHC and quantification of ESR1, PGR and MIKI67RNA expression with RT-qPCR both represent promis-ing alternatives to conventional visual estimation andmay assist in improving reproducibility and accuracy inthe field. However, RT-qPCR assessment of tumor pro-liferation was overall more accurate than quantitativeIHC. This is the first study to compare tumor MKI67gene expression by RNA and protein assessment in aprospective retrospective neoadjuvant setting. Due tothe relatively small sample size, these data should beconsidered preliminary and worth validating in largerdatasets.

Sinn et al. BMC Cancer (2017) 17:124 Page 8 of 10

Additional file

Additional file 1: Raw data; qPCR data, quantitative and visual IHCvalues, pathologic complete response yes/no. (XLSX 42 kb)

AbbreviationsCISH: Chromogenic in situ hybridization; Cq: Quantification cycle;DDFS: Distant disease free survival; ESR1/ER: Oestrogen receptor alpha;ERBB2/HER2; FEC: Fluorouracil & epirubicin, cyclophosphamidechemotherapy; FFPE: Formalin fixed paraffin embedded; GOI: Gene ofinterest; HR: Hazard ratio; IHC: Immunohistochemistry; MKI67/Ki67: markerof proliferation Ki-67; mRNA: Messenger ribonucleic acid; NPA: Negativepercentage agreement; OPA: Overall percentage agreement; OS: Overallsurvival; PGR/PgR: Progesterone receptor; PPA: Positive percentageagreement; REF: Reference gene; RT-qPCR: Reverse transcriptionquantitative real time polymerase chain reaction

AcknowledgementsWe thank Susanne Scharff, Silke Claas and Torsten Acht for excellent technicalsupport in developing molecular subtyping technologies, and Drs. ThomasKeller and Stefan Weber for performing statistical analyses.We acknowledge the financial support of the DeutscheForschungsgemeinschaft and Ruprecht-Karls-Universität Heidelberg withinthe funding programme Open Access Publishing.

FundingDietmar Hopp-Stiftung,Award Number: 23011195,Grant Recipient: Peter Sinn MD PhD

Availability of data and materialsThe dataset has been made available in the Excel file format as Additional file 1.

Authors’ contributionsHPS, KS, ML, SL, RMW and FM were involved in the conception of the study. EVperformed the RT-qPCR assays; HPS, AS, MK, JS and FM provided study dataand materials; HPS, SL, PA and RMW performed the statistical analysis and wrotethe statistical plan; HPS, AS, KS, ML, SL, SE, RMW and FM interpreted the data;HPS, SL, RMW and FM drafted the manuscript; all authors read and approvedthe final manuscript.

Competing interestsRMW and SE are founders of STRATIFYER Molecular Pathology GmbH. SL, EVand RMW are employees of STRATIFYER Molecular Pathology GmbH. KS andML are employees of BioNTech Diagnostics GmbH. All other authors declarethat they have no competing interests.

Consent for publicationNot applicable.

Ethics approval and consent to participateAll patients had been enrolled in the H3E-MC-S080 (NCT00149214) neoadjuvantphase II study [1] at a single institution (National Center for Tumor Diseases,University-Hospital, Heidelberg). A written informed consent for the researchuse of patient biological material was granted at the time of enrolment. Theresearch described herein is completely independent from the sponsor of theoriginal study (Eli Lilly and Company). The study was approved by the localethics committee of the Heidelberg University Hospital.

Author details1Institute of Pathology, University of Heidelberg, Im Neuenheimer Feld220-221, 69120 Heidelberg, Germany. 2National Center for Tumor Diseases,University-Hospital Heidelberg, Im Neuenheimer Feld 460, 69120 Heidelberg,Germany. 3German Cancer Research Center, Im Neuenheimer Feld 280,69120 Heidelberg, Germany. 4STRATIFYER Molecular Pathology GmbH,Werthmannstr. 1c, 50935 Köln, Germany. 5Department of Pathology, St.Elisabeth-Krankenhaus, Werthmannstr. 1c, 50935 Köln, Germany. 6BioNTechDiagnostics GmbH, 55131 Mainz, Germany.

Received: 8 April 2016 Accepted: 4 February 2017

References1. Senkus E, Kyriakides S, Ohno S, Penault-Llorca F, Poortmans P, Rutgers E,

Zackrisson S, Cardoso F, Committee EG. Primary breast cancer: ESMOclinical practice guidelines for diagnosis, treatment and follow-up. AnnOncol. 2015;26 Suppl 5:v8–30.

2. Coates AS, Winer EP, Goldhirsch A, Gelber RD, Gnant M, Piccart-Gebhart M,Thurlimann B, Senn HJ, Panel M. Tailoring therapies–improving themanagement of early breast cancer: St Gallen International ExpertConsensus on the Primary Therapy of Early Breast Cancer 2015. Ann Oncol.2015;26(8):1533–46.

3. Gerdes J, Lemke H, Baisch H, Wacker HH, Schwab U, Stein H. Cell cycleanalysis of a cell proliferation-associated human nuclear antigen defined bythe monoclonal antibody Ki-67. J Immunol. 1984;133(4):1710–5.

4. Braun L, Mietzsch F, Seibold P, Schneeweiss A, Schirmacher P, Chang-ClaudeJ, Sinn P, Aulmann S. Intrinsic breast cancer subtypes defined by estrogenreceptor signalling-prognostic relevance of progesterone receptor loss. ModPathol. 2013;26(9):1161–71.

5. Maisonneuve P, Disalvatore D, Rotmensz N, Curigliano G, Colleoni M,Dellapasqua S, Pruneri G, Mastropasqua MG, Luini A, Bassi F, et al. Proposednew clinicopathological surrogate definitions of luminal A and luminal B(HER2-negative) intrinsic breast cancer subtypes. Breast Cancer Res.2014;16(3):R65.

6. Peppercorn J, Perou CM, Carey LA. Molecular subtypes in breast cancerevaluation and management: divide and conquer. Cancer Invest.2008;26(1):1–10.

7. Hammond ME, Hayes DF, Dowsett M, Allred DC, Hagerty KL, Badve S,Fitzgibbons PL, Francis G, Goldstein NS, Hayes M, et al. American Societyof Clinical Oncology/College of American Pathologists guidelinerecommendations for immunohistochemical testing of estrogen andprogesterone receptors in breast cancer. Arch Pathol Lab Med. 2010;134(6):907–22.

8. de Gramont A, Watson S, Ellis LM, Rodon J, Tabernero J, de Gramont A,Hamilton SR. Pragmatic issues in biomarker evaluation for targetedtherapies in cancer. Nat Rev Clin Oncol. 2015;12(4):197–212.

9. Harvey J, Clark G, Osborne C, Allred D. Estrogen receptor status byimmunohistochemistry is superior to the ligand-binding assay forpredicting response to adjuvant endocrine therapy in breast cancer.J Clin Oncol. 1999;17(5):1474–81.

10. Luporsi E, Andre F, Spyratos F, Martin PM, Jacquemier J, Penault-Llorca F,Tubiana-Mathieu N, Sigal-Zafrani B, Arnould L, Gompel A, et al. Ki-67: levelof evidence and methodological considerations for its role in the clinicalmanagement of breast cancer: analytical and critical review. Breast CancerRes Treat. 2012;132(3):895–915.

11. Polley MY, Leung SC, McShane LM, Gao D, Hugh JC, Mastropasqua MG,Viale G, Zabaglo LA, Penault-Llorca F, Bartlett JM, et al. An international Ki67reproducibility study. J Natl Cancer Inst. 2013;105(24):1897–906.

12. Varga Z, Diebold J, Dommann-Scherrer C, Frick H, Kaup D, Noske A,Obermann E, Ohlschlegel C, Padberg B, Rakozy C, et al. How reliable is Ki-67immunohistochemistry in grade 2 breast carcinomas? A QA study of theSwiss working group of breast- and gynecopathologists. PLoS ONE. 2012;7(5):e37379.

13. Yerushalmi R, Woods R, Ravdin PM, Hayes MM, Gelmon KA. Ki67 in breastcancer: prognostic and predictive potential. Lancet Oncol. 2010;11(2):174–83.

14. Denkert C, Loibl S, Muller BM, Eidtmann H, Schmitt WD, Eiermann W, Gerber B,Tesch H, Hilfrich J, Huober J, et al. Ki67 levels as predictive and prognosticparameters in pretherapeutic breast cancer core biopsies: a translationalinvestigation in the neoadjuvant GeparTrio trial. Ann Oncol. 2013;24(11):2786–93.

15. Di Cataldo S, Ficarra E, Macii E. Computer-aided techniques forchromogenic immunohistochemistry: status and directions. Comput BiolMed. 2012;42(10):1012–25.

16. Voros A, Csorgo E, Nyari T, Cserni G. An intra- and interobserverreproducibility analysis of the Ki-67 proliferation marker assessment on corebiopsies of breast cancer patients and its potential clinical implications.Pathobiology. 2013;80(3):111–8.

17. Konsti J, Lundin M, Joensuu H, Lehtimaki T, Sihto H, Holli K, Turpeenniemi-Hujanen T, Kataja V, Sailas L, Isola J, et al. Development and evaluation of avirtual microscopy application for automated assessment of Ki-67expression in breast cancer. BMC Clin Pathol. 2011;11:3.

Sinn et al. BMC Cancer (2017) 17:124 Page 9 of 10

18. Cass JD, Varma S, Day AG, Sangrar W, Rajput AB, Raptis LH, Squire J,Madarnas Y, Sengupta SK, Elliott BE. Automated quantitative analysis of p53,Cyclin D1, Ki67 and pERK expression in breast carcinoma does not differfrom expert pathologist scoring and correlates with clinico-pathologicalcharacteristics. Cancers (Basel). 2012;4(3):725–42.

19. Abubakar M, Howat WJ, Daley F, Zabaglo L, McDuffus LA, Blows F, CoulsonP, Raza Ali H, Benitez J, Milne R, et al. High-throughput automated scoringof Ki67 in breast cancer tissue microarrays from the Breast CancerAssociation Consortium. J Pathol Clin Res. 2016;2(3):138–53.

20. Noske A, Loibl S, Darb-Esfahani S, Roller M, Kronenwett R, Muller BM, SteffenJ, von Toerne C, Wirtz R, Baumann I, et al. Comparison of differentapproaches for assessment of HER2 expression on protein and mRNA level:prediction of chemotherapy response in the neoadjuvant GeparTrio trial(NCT00544765). Breast Cancer Res Treat. 2011;126(1):109–17.

21. Bohmann K, Hennig G, Rogel U, Poremba C, Mueller BM, Fritz P, Stoerkel S,Schaefer KL. RNA extraction from archival formalin-fixed paraffin-embeddedtissue: a comparison of manual, semiautomated, and fully automatedpurification methods. Clin Chem. 2009;55(9):1719–27.

22. Laible M, Schlombs K, Kaiser K, Veltrup E, Herlein S, Lakis S, Stohr R, Eidt S,Hartmann A, Wirtz RM, et al. Technical validation of an RT-qPCR in vitrodiagnostic test system for the determination of breast cancer molecularsubtypes by quantification of ERBB2, ESR1, PGR and MKI67 mRNA levelsfrom formalin-fixed paraffin-embedded breast tumor specimens. BMCCancer. 2016;16:398.

23. Elsawaf Z, Sinn HP, Rom J, Bermejo JL, Schneeweiss A, Aulmann S. Biologicalsubtypes of triple-negative breast cancer are associated with distinctmorphological changes and clinical behaviour. Breast. 2013;22(5):986–92.

24. Pepe MS, Janes H, Longton G, Leisenring W, Newcomb P. Limitations of theodds ratio in gauging the performance of a diagnostic, prognostic, orscreening marker. Am J Epidemiol. 2004;159(9):882–90.

25. Cook NR. Use and misuse of the receiver operating characteristic curve inrisk prediction. Circulation. 2007;115(7):928–35.

26. Guarneri V, Broglio K, Kau SW, Cristofanilli M, Buzdar AU, Valero V, BuchholzT, Meric F, Middleton L, Hortobagyi GN, et al. Prognostic value of pathologiccomplete response after primary chemotherapy in relation to hormonereceptor status and other factors. J Clin Oncol. 2006;24(7):1037–44.

27. Cortazar P, Geyer Jr CE. Pathological complete response in neoadjuvanttreatment of breast cancer. Ann Surg Oncol. 2015;22(5):1441–6.

28. Berruti A, Generali D, Kaufmann M, Puztai L, Curigliano G, Aglietta M, GianniL, Miller WR, Untch M, Sotiriou C, et al. International expert consensus onprimary systemic therapy in the management of early breast cancer:highlights of the fourth symposium on primary systemic therapy in themanagement of operable breast cancer, Cremona, Italy (2010). J Natl CancerInst Monogr. 2011;2011(43):147–51.

29. Mauri D, Pavlidis N, Ioannidis JP. Neoadjuvant versus adjuvant systemictreatment in breast cancer: a meta-analysis. J Natl Cancer Inst.2005;97(3):188–94.

30. Fasching PA, Heusinger K, Haeberle L, Niklos M, Hein A, Bayer CM, Rauh C,Schulz-Wendtland R, Bani MR, Schrauder M, et al. Ki67, chemotherapyresponse, and prognosis in breast cancer patients receiving neoadjuvanttreatment. BMC Cancer. 2011;11:486.

31. Cheang MC, Chia SK, Voduc D, Gao D, Leung S, Snider J, Watson M, Davies S,Bernard PS, Parker JS, et al. Ki67 index, HER2 status, and prognosis of patientswith luminal B breast cancer. J Natl Cancer Inst. 2009;101(10):736–50.

32. von Minckwitz G, Untch M, Blohmer JU, Costa SD, Eidtmann H, Fasching PA,Gerber B, Eiermann W, Hilfrich J, Huober J, et al. Definition and impact ofpathologic complete response on prognosis after neoadjuvantchemotherapy in various intrinsic breast cancer subtypes. J Clin Oncol. 2012;30(15):1796–804.

33. Potemski P, Pluciennik E, Bednarek AK, Kusinska R, Kubiak R, Jesionek-Kupnicka D, Watala C, Kordek R. Ki-67 expression in operable breast cancer:a comparative study of immunostaining and a real-time RT-PCR assay.Pathol Res Pract. 2006;202(7):491–5.

34. Yamamoto S, Ibusuki M, Yamamoto Y, Fu P, Fujiwara S, Murakami K, Iwase H.Clinical relevance of Ki67 gene expression analysis using formalin-fixedparaffin-embedded breast cancer specimens. Breast Cancer. 2013;20(3):262–70.

35. Niikura N, Iwamoto T, Masuda S, Kumaki N, Xiaoyan T, Shirane M, Mori K,Tsuda B, Okamura T, Saito Y, et al. Immunohistochemical Ki67 labeling indexhas similar proliferation predictive power to various gene signatures inbreast cancer. Cancer Sci. 2012;103(8):1508–12.

36. Klauschen F, Wienert S, Schmitt WD, Loibl S, Gerber B, Blohmer JU, HuoberJ, Rudiger T, Erbstosser E, Mehta K, et al. Standardized Ki67 diagnostics usingautomated scoring–clinical validation in the GeparTrio breast cancer study.Clin Cancer Res. 2015;21(16):3651–7.

37. Mohammed ZM, McMillan DC, Elsberger B, Going JJ, Orange C, Mallon E,Doughty JC, Edwards J. Comparison of visual and automated assessment ofKi-67 proliferative activity and their impact on outcome in primary operableinvasive ductal breast cancer. Br J Cancer. 2012;106(2):383–8.

38. Gianni L, Zambetti M, Clark K, Baker J, Cronin M, Wu J, Mariani G, RodriguezJ, Carcangiu M, Watson D, et al. Gene expression profiles in paraffin-embedded core biopsy tissue predict response to chemotherapy in womenwith locally advanced breast cancer. J Clin Oncol. 2005;23(29):7265–77.

39. Parker JS, Mullins M, Cheang MC, Leung S, Voduc D, Vickery T, Davies S,Fauron C, He X, Hu Z, et al. Supervised risk predictor of breast cancer basedon intrinsic subtypes. J Clin Oncol. 2009;27(8):1160–7.

40. Straver ME, Glas AM, Hannemann J, Wesseling J, van de Vijver MJ, RutgersEJ, Vrancken Peeters MJ, van Tinteren H, Van't Veer LJ, Rodenhuis S. The 70-gene signature as a response predictor for neoadjuvant chemotherapy inbreast cancer. Breast Cancer Res Treat. 2010;119(3):551–8.

41. Cuzick J, Dowsett M, Pineda S, Wale C, Salter J, Quinn E, Zabaglo L, Mallon E,Green AR, Ellis IO, et al. Prognostic value of a combined estrogen receptor,progesterone receptor, Ki-67, and human epidermal growth factor receptor 2immunohistochemical score and comparison with the Genomic Healthrecurrence score in early breast cancer. J Clin Oncol. 2011;29(32):4273–8.

42. Tobin NP, Harrell JC, Lovrot J, Egyhazi Brage S, Frostvik Stolt M, Carlsson L,Einbeigi Z, Linderholm B, Loman N, Malmberg M, et al. Molecular subtypeand tumor characteristics of breast cancer metastases as assessed by geneexpression significantly influence patient post-relapse survival. Ann Oncol.2015;26(1):81–8.

43. Williams DJ, Cohen C, Darrow M, Page AJ, Chastain B, Adams AL.Proliferation (Ki-67 and phosphohistone H3) and oncotype DX recurrencescore in estrogen receptor-positive breast cancer. Appl ImmunohistochemMol Morphol. 2011;19(5):431–6.

• We accept pre-submission inquiries

• Our selector tool helps you to find the most relevant journal

• We provide round the clock customer support

• Convenient online submission

• Thorough peer review

• Inclusion in PubMed and all major indexing services

• Maximum visibility for your research

Submit your manuscript atwww.biomedcentral.com/submit

Submit your next manuscript to BioMed Central and we will help you at every step:

Sinn et al. BMC Cancer (2017) 17:124 Page 10 of 10